Privacy as Protection of the Incomputable Self: From Agnostic to Agonistic Machine Learning

“There has recently been a big legal and political debate on the extent to which the introduction of AI-based automated (administrative) decision-making is desirable and legitimate (European Commission, 2019; Hildebrandt, 2019; Noto La Diega, 2018).”

“Wellbeing, in the context of AI governance, refers to the overall quality of life and flourishing of individuals and society, encompassing aspects such as physical and mental health, safety, privacy, autonomy, fairness, and social cohesion (Floridi et al., 2018; IEEE Global Initiative on Ethics of Autonomous Intelligen…”

“As Hildebrandt (2019) advocates, societies must demand companies to explore and enable alternative ways of datafying and modeling the same person.”

“Yet, the more general idea of “agonistic machine learning” can be desirable even if one does not buy the additional premises adopted by Hilde-brandt [15], as it would subject modelling and processing choices to internal and external evaluation regarding their compliance to the standards set by…”

“In-deed, Hildebrandt has highlighted that data protection requires“thejustificationofsuchdecision-making rather than an explanation in the sense of its heuristics” (p. 113 in [18]).”

“…purposes—a technology entering a sphere of life and intruding on privacy—GSM technology makes it possible to ‘view’ an individual’s movements because the environment has been reconfigured by sensor and networked technologies to capture a person’s movements (Hildebrandt, 2019, pp. 89–95).”

“The distinction between the two notions is best illustrated by Mireille Hildebrandt (2019): “If the decisions of an automated machine learning application indirectly discriminate on the basis of gender or race they may qualify as prohibited discrimination; explaining why the system so decided may…”

“40 It seems that human participation is required of decision-makers and of those subjected to the consequences of decisions.”

“The following quote from Mark Zuckerberg is illustrative: “I’m also curious about whether there is a fundamental mathematical law underlying human social relationships that governs the balance of who and what we all care about [...] I bet there is” [Hildebrandt, 2019].”

“Similarly, adversarialism is essential to the rule of law (Hildebrandt, 2019).”

“In contrast to the inductive bias — the form of bias required by an algorithm to learn patterns (Hildebrandt, 2019) — such a bias can impair the generalizability of a hate speech detection model (Wiegand et al., 2019; Geva et al., 2019) or can lead to unfair models (e.g., discriminating minorities)…”

“The same holds at the level of human learning, as discussed in the area of philosophical hermeneutics [23].”

“Steps towards countering testimonial injustice include: considering whether platforms such as Twitter are a sufficient source of evidence for understanding population beliefs, preferences and behaviors; and adopting ‘agonistic’ machine learning, where ‘companies or governments that base decisions on machine learning m…”

“After a series of operations such as data collection, preprocessing, and feature extraction, a set of feature vectors describing basketball actions can be obtained, and the classification to which these abstract features belong can be obtained using the classifier model (Hildebrandt, 2019).”

“11 This must however take into account the limitations in Article 19(3) and unlawful forms of expression prohibited under Article 20 ICCPR. modulated through the widespread use of digital technologies such as AI [ 76, 77].”

“First, concerns around dignity and individual rights such as autonomy are relevant [4, 32, 60].”

“Rapid developments in networked technologies pose an intellectual freedom conundrum: that as technology provides users with increasing access to ideas and affordances for extended cognition, it simultaneously renders their internal mental processes subject to observation and influence (Cohen, 1996; Gandy, 1993; Hildeb…”

“Instead, we should embrace the incomplete and the incomputable (Hildebrandt, 2019).”

“Hildebrandt sets out that ‘in real life situations … future data can always disrupt the predictive accuracy of the hypothesis target function’ (Hildebrandt 2019).”

“It is knowledge that relegates any other facets from the perception of the self, facets which can only be manifested through language (Hildebrandt 2019; Rouvroy 2013; Rouvroy and Stiegler 2016).”

“…of ‘free will and deliberation’, the challenge now, in tune with Arendt’s critique on behavioralist modes of governance, is ‘to protect the incomputable nature of the human self, its foundational indeterminacy and the natality it expresses’ (Hildebrandt 2016c, p. 29; Hildebrandt 2019, p. 105).”

“Scholarship has pointed out many dangers such as risks of discrimination [3, 20, 28, 54], violations of privacy [5, 17, 21, 28, 31, 38, 52, 57]”

“As for advanced analytics (Mayer-Schönberger and Padova, 2016; Rouvroy, 2016; De Raedt, 2017; Zarsky, 2017; Hildebrandt, 2019), the opportunities offered by fraud analytics must be balanced with the need to protect the citizens ’ right to privacy and personal data protection (De Raedt, 2017;…”

“Theoretical arguments have illustrated how contestability can help promote fairness and accountability in decision-making [39, 41, 45], protect individual rights [10, 39, 45], and preserve human dignity and autonomy [50].”

“…and AI development could help to “call out the ethical and political implications of who decides task T , performance metric P and experience E , and to investigate how this is done , taking into account which (and whose) concerns are at stake ” (Hildebrandt, 2019, p. 110, cursive in the original).”

“The separation of the ‘real’ and the ‘digital’ leads to conceptions of privacy and data protection that focus on control and the flow of data, instead of the individual’s right to form and construct their own identity—“to protect what counts but cannot be counted ” (Hildebrandt, 2019 , p. 96).”

“The concept of privacy likewise concerns individuals’ ability to influence the construction of their own identity (Søe & Mai, 2022) and is as such a right to protect human subjectivity (Hildebrandt, 2019).”

“[48, 49, 52] argued that explanations are not sufcient to justify a decision.”

“Legal scholars have framed human intervention as a means to protect decision-subjects’ fundamental right of human dignity [4]; a measure to acknowledge the “foundational indeterminacy of human self” [48, 73].”

“There is growing recognition in user profiling and user modeling communities that such profiling techniques create unique ethical challenges [3, 30, 63].”

“Government guidelines will help policymakers redress volatile biases in data [53].”

“…does not do justice to “the incomputable self” as the origin of the ambiguous nature of human inter-relations, born out of “facing the uncertainty of being (mis)understood in one way or another “, which Hildebrandt characterises as the very indeterminacy where human freedom is situated [41, p. 89].”

“Providing human review of decisions made using algorithms that significantly impact a person’s life, is arguably a way to “uphold human dignity” [5, 29, 41].”

“The attention that GDPR gives to privacy-related concerns, particularly those inherent in ML techniques, has led some to place high confidence in the regulation [102].”

“However, since value reasoning is cognitively challenging [19, 29] and implicit in human thinking [16, 21], value inference cannot be performed solely via computational methods.”

“Beneath the performance of identification, dissolution and fuzzy boundaries that constitute our identities—into the importance of the uncountable and incomputable (Hildebrandt, 2019)—we must therefore”

“It furthermore creates a documentation of the intended behavior of the ML model that serves as a pre hoc explanation and justification, similar to the preregistra-tion of an experimental setup [19].”

“It can be argued as a matter of normative ethics that human beings might lose their sense of autonomy necessary for living a good life if machines increasingly influence and guide human activities (Hildebrandt, 2019).”

“As scholars such as Hildebrandt [24] and Noble [25] have demonstrated, algorithmic systems inherit the biases of their creators, the structural inequalities of their data, and the political rationalities embedded in their design.”

“Institutional memory: Building on collective learning about uncertainty handling These principles align with what (Hildebrandt, 2019 ) calls ‘agonistic machine learn - ing’—approaches that recognize and preserve rather than resolve tensions around uncertainty interpretation.”

“Beyond the widely discussed issues of privacy in data collection for consumers from the social sciences [5, 58] and computer science [43], these data supply chains also involve a different array of actors, including data workers [75], data brokers [4], and other institutional and governmental actors.”

“…are incom - patible with procedural fairness if the use of these systems prevents institutions from recognising and accounting for the unique positionality, capabilities, and needs of people affected by their decision, in reducing them to datapoints, for example (Dao, 2020; Hildebrandt, 2019b ).”